CN114707396A - All-time PM2.5Near real-time production method of concentration seamless lattice point data - Google Patents
All-time PM2.5Near real-time production method of concentration seamless lattice point data Download PDFInfo
- Publication number
- CN114707396A CN114707396A CN202111625657.3A CN202111625657A CN114707396A CN 114707396 A CN114707396 A CN 114707396A CN 202111625657 A CN202111625657 A CN 202111625657A CN 114707396 A CN114707396 A CN 114707396A
- Authority
- CN
- China
- Prior art keywords
- data
- time
- aod
- concentration
- seamless
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Geometry (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a full-time PM2.5A near real-time production method of concentration seamless lattice point data relates to the field of atmospheric environment remote sensing monitoring, and comprises the following steps: inverting the daytime hourly aerosol optical thickness; reconstructing missing values of the AOD data; correcting errors of the aerosol simulation result in the numerical mode; establishing a machine learning model of the atmospheric pollutant concentration observation data actually measured by the national control station and AOD; performing multi-scale AOD fusion analysis; estimating near-surface PM2.5Concentration; construction of PM2.5A spatiotemporal migration model of concentration; fusing site actual measurement and estimation PM2.5Concentration data to produce hourly seamless PM2.5Grid point data; the invention can obtain the whole dayTime-hour resolution and spatially seamless PM2.5Concentration grid data, in particular, surface area PM under the condition of realizing non-satellite inversion AOD data2.5The key technology for the rapid production of concentration data.
Description
Technical Field
The invention relates to the field of atmospheric environment remote sensing monitoring, in particular to a full-time PM2.5A near real-time production method of concentration seamless lattice point data.
Background
In recent years, PM is used2.5The dominant pollution of atmospheric particulates has been a major factor affecting urban air quality due to PM2.5Small particle size, large specific surface area and easy long-term detention in the air, thus having serious threat to public health and global ecological environment. Therefore, near real-time monitoring of near-surface PM2.5The concentration has important practical significance for realizing the fine management and control of atmospheric particulate pollution and the prevention and control of dust haze. However, due to sparse and uneven distribution of sites, a large-area monitoring blind area still exists in the existing ground-based air quality observation network. Meanwhile, more manpower and material resources are needed for site business operation. Therefore, the existing site observation data are difficult to satisfy the area PM2.5And the concentration can be comprehensively tracked and monitored.
Research shows that a statistical model based on data driving can better depict the mapping relation between the atmospheric Aerosol and the atmospheric particulate matter concentration, so that the satellite remote sensing is utilized to invert the Optical thickness (AOD) data of the Aerosol to develop the near-ground PM2.5Concentration estimation, which has become the current quantitative acquisition region PM2.5The important technical means of concentration. However, due to interference of bright earth surfaces such as cloud and snow, large-area data loss often exists in the satellite inversion AOD product, and meanwhile, quantitative monitoring of atmospheric aerosol parameters at night is difficult to be considered by the AOD inversion algorithm based on the radiation transmission theory, so that the estimation is accordingly carried outCalculated PM2.5The concentration lattice point data cannot realize 24-h space-time full coverage. In addition, although the numerical simulation result based on the atmospheric chemical transmission mode can provide spatio-temporal continuous aerosol parameter reanalysis data, the spatial resolution of the product is low and the result has great uncertainty due to the simplification of the mode simulation process and the lack of key basic data such as a near-real-time discharge list.
In view of the above problems, there is a need for a technical solution capable of synchronously solving the following technical problems:
(1) satellite remote sensing inversion and missing information reconstruction of near real-time AOD data in daytime;
(2) area PM under the condition of lack of AOD observation such as night2.5Estimating the concentration;
(3) performing deviation correction and fusion analysis among multi-source heterogeneous data;
to achieve near real-time PM2.5Efficient production of seamless grid point data with concentration meeting regional PM2.5The requirement of all-time and all-direction tracking and monitoring of pollution.
Disclosure of Invention
The invention aims to provide a full-time PM2.5A near real-time production method of concentration seamless lattice point data is provided to solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme: all-time PM2.5The near real-time production method of the concentration seamless lattice point data comprises the following steps:
step 1: rapidly inverting day hour-by-hour AOD data by depending on a remote sensing observation platform of a static satellite;
and 2, step: reconstructing missing information in the AOD lattice point data produced in the step 1;
and 3, step 3: performing spatial downscaling and error correction on the AOD result of the numerical mode simulation based on satellite inversion AOD data by using a machine learning algorithm;
and 4, step 4: estimating the AOD data level of the corresponding point by utilizing the actually measured atmospheric particulate matter concentration data on the ground in combination with a random forest machine learning modeling method;
and 5: taking the numerically-simulated AOD result obtained in the step 3 after the size reduction as a background field, and fusing the multi-source AOD data obtained in the step 2 and the step 4 to obtain a daytime high-precision space seamless AOD product;
step 6: utilizing the seamless AOD product obtained in the step 5 and combining with the ground state control station to actually measure PM2.5Concentration, real-time meteorological observation data, social and economic data and time variable related to atmospheric pollution are modeled by adopting a random forest machine learning method to estimate surface area PM2.5Concentration distribution data;
and 7: aiming at the condition that no effective satellite remote sensing inverses AOD data at night and the like, PM between different moments is constructed2.5The concentration migration model realizes PM of other time areas based on the existing data of the adjacent time2.5Estimating concentration data;
and 8: developing actual measurement PM of site at current moment2.5Concentration data and space-time seamless PM obtained in step 6 or 72.5Integration between lattice data products, high-precision full-coverage PM at current production time2.5The density grid data.
As a preferred technical scheme of the invention, the step 1 depends on the high time-frequency observation characteristic of the geostationary satellite, and adopts a minimum wave spectrum regression coefficient method to estimate the surface reflectivity so as to realize accurate ground gas decoupling.
As a preferred technical solution of the present invention, in the step 1, based on the time-space autocorrelation characteristics of the AOD data, an empirical orthogonal function method is used to fuse multi-temporal incomplete AOD data to realize reconstruction of partial missing data, so as to improve the spatial coverage level of the satellite inversion AOD raw data obtained in the step 1.
As a preferred technical solution of the present invention, the data reconstruction technique adopted in step 2 is specifically described as introducing the AOD data of the spatio-temporal neighborhood, constructing the AOD spatio-temporal correlation matrix at the current time, performing iterative decomposition on the AOD spatio-temporal correlation matrix by using a Singular Value Decomposition (SVD), and iteratively reconstructing missing values in the spatio-temporal correlation matrix by using the main mode until convergence.
As a preferred technical solution of the present invention, the data of the concentration of the ground actual measurement atmospheric particulates in step 4 is the real-time air pollutant concentration provided by the national control air quality observation network, and the statistical modeling formula for estimating the AOD in step 4 is as follows:
AOD～PM10+PM2.5+NO2+SO2+MET
wherein the AOD data is provided from an atmospheric aerosol characterization observation site, PM10、PM2.5、NO2And SO2The concentration data is provided by a national foundation air quality monitoring network, and MET represents meteorological observation data at a corresponding moment.
As a preferred technical solution of the present invention, in the step 5, an optimal interpolation fusion technique is used to perform fusion analysis on the mentioned multi-source heterogeneous AOD data:
xa＝xb+K(Yo-HXb)
wherein x isaTo re-analyze the value, xbFor background values, K is the Kalman gain, YoIs the observed value in the x neighborhood of the pixel to be analyzed, H is the observation operator, HXbRepresents an observed value YoCorresponding background values. In the invention, the reduced-scale mode simulation result is used as a fusion background field, and the satellite inversion result and the ground estimation data are used as a fusion observation field.
As a preferred technical solution of the present invention, the systematic error correction of the AOD products with different data sources is represented by the corresponding variance of the errors of the background field and the observation field:
and epsilon is var (X-Y), wherein X is measured data, and Y is data to be calibrated.
As a preferred technical solution of the present invention, the modeling formula in step 6 is:
PM2.5～AOD+MET+SE+TIME
wherein PM2.5The data is provided by a national foundation air quality monitoring network, the AOD uses the space-TIME full coverage reanalysis data produced in the step 5, MET represents meteorological observation data at a corresponding moment, SE represents social and economic data related to atmospheric pollution, and TIME represents a corresponding TIME variable.
In a preferred embodiment of the present invention, in the step 7, PM is determined between different times2.5The concentration migration model is constructed by a deep neural network learning model, and the expression of the deep neural network learning model is as follows:
wherein t represents the current time, t-m represents the forward m-time with the t-time as the reference, LSTM is the long-term and short-term memory layer, FC is the full-link layer, the migration model is constructed based on the continuous measured data of the sites, and model extrapolation is applied to realize the surface domain scale PM2.5And (4) estimating the concentration.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a full-time PM2.5The near-real-time production method of seamless grid point data of concentration is focused on realizing 24-h space-time continuous PM2.5And (3) production of concentration grid point data to solve the problem of data shortage of all-round tracking monitoring of atmospheric particulate pollution all day long. Compared with other similar technical inventions, the production technology provided by the invention can obtain PM with small time and small resolution all day and seamless space2.5Concentration grid point data, especially realizing surface area PM under the condition of non-satellite inversion AOD data at night2.5The key technology for the rapid production of concentration data. The invention can solve the problem of seamless high-resolution PM of the related industries on space-time2.5The actual application requirements of the concentration data.
Drawings
FIG. 1 is a full time PM of the present invention2.5A schematic flow diagram of a near real-time production method of concentration seamless grid data;
FIG. 2 is a flow chart of an algorithm for remote sensing inversion of AOD data by a geostationary satellite according to the present invention;
FIG. 3 is a schematic diagram of a multi-source multi-scale AOD data fusion analysis in the present invention;
FIG. 4 shows a PM according to the present invention2.5And (5) verifying the scatter diagram of the concentration seamless grid point data.
Detailed Description
For those skilled in the art to better understand the technical solutions of the embodiments of the present application, the following will make clear and complete descriptions of the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, not all the embodiments; based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application;
it should be noted that, in the present application, the embodiments and features of the embodiments may be combined with each other without conflict, and the present application will be described in detail with reference to fig. 1 to 4 in conjunction with the embodiments;
the invention provides a full-time PM2.5Near real-time production method of seamless grid point data of concentration, solving current hourly PM2.5The key technical problem of the production of concentration seamless lattice point data, in particular to the related technologies of atmospheric parameter inversion, missing information reconstruction, deep learning modeling, data fusion assimilation and the like, as shown in figure 1, the method comprises the following steps:
step 1: according to the high time frequency observation characteristic of the geostationary satellite, the invention provides a high-precision AOD fast inversion algorithm to realize the real-time production of day AOD lattice point data.
a) Acquiring pixel values of visible light, near infrared, short wave infrared, thermal infrared and other wave bands in real time through a static satellite data release platform, and carrying out radiometric calibration on data of each wave band according to a calibration coefficient to generate satellite apparent reflectivity or bright temperature data with physical significance;
b) reading observation geometric parameters such as a solar zenith angle, an azimuth angle, a sensor zenith angle, an azimuth angle and the like from the geographic positioning data, and calculating concrete longitude and latitude information of each pixel according to a row and column number and longitude and latitude conversion relation given by a national satellite meteorological center website;
c) according to the differences of the reflectivity and the bright temperature difference of different types of pixels in visible light, short wave infrared and thermal infrared bands, a multiband combined threshold value method is adopted to quickly detect cloud, ice and snow and inland water body pixels;
d) removing masks of pixels with the sun zenith angle or the observation zenith angle larger than 75 degrees to eliminate the increment of uncertainty of inversion results caused by overlarge zenith angle;
c) the surface reflectivity is modeled and solved by using a minimum spectrum regression coefficient method (taking FY-4A geostationary satellite as an example), and specifically:
virtual earth surface reflectivities in the 0.47, 0.65, and 2.25 μm bands in a clean background (e.g., AOD ═ 0.05) were calculated by simulation based on the 6SV radiation transmission mode using an atmospheric correction module.
Using blue band 0.47 μm as an example, define the variables
The minimum value of the ratio of the blue light to the short wave infrared band virtual earth surface reflectivity in a given time window T, namely the minimum spectrum regression coefficient:
because the real aerosol in a certain time interval in the time window T is higher than the set background aerosol level, the contribution of the partial apparent reflectivity of the blue light wave interval is included in the virtual surface reflectivity, and the result is that
Overestimating. Therefore, the temperature of the molten metal is controlled,
the ratio of the real surface reflectivity equivalent to two bands can be approximated.
According to the 2.25 μm band, the aerosol hardly affects the particle diameter, so that
Can be approximated to
Thus, the surface reflectivity of the blue band at time t
Can pass through
And
is estimated as the product of:
the method has universality to most earth surface types because each pixel is independently calculated;
d) according to the characteristics of the aerosol observed by the atmospheric aerosol ground station, inducing the particle spectrum distribution and the complex refractive index of different months in China to generate a local typical aerosol mode in China;
e) calculating the apparent reflectivities of blue light, red light and short wave infrared bands under different earth surface reflectivities and AOD levels by using a 6SV radiation transmission mode and based on the drawn-up solar and satellite geometric parameters, an atmospheric mode, an aerosol mode and a satellite spectral response function, and constructing AOD inversion lookup tables of different aerosol modes;
f) and developing AOD fast interpolation inversion by utilizing the constructed lookup table according to the real-time observation data, the geometric parameters and the estimated earth surface reflectivity of each pixel.
Step 2: and (3) reconstructing the AOD data missing value obtained by the satellite remote sensing inversion in the step (1), and improving the spatial coverage level of the AOD lattice point data.
a) In a given space range, introducing historical AOD data in a certain time window, and constructing an AOD space-time correlation matrix X based on the current small-time resolution:
where m and n represent the number of rows and columns of the AOD image in the spatial domain, t0Representing the current time to be reconstructed, tkRepresenting k historical moments;
b) reserving 10-20% of effective values in the original correlation matrix as actual measurement data for judging reconstruction accuracy, and filling a time domain mean value as an initial value into a matrix vacancy position to initialize EOF analysis;
c) the space-time correlation matrix is decomposed into the following three parts by using an SVD method:
[U，S，V]＝SVD(X)
the method comprises the following steps that U and V respectively represent time and space modes, S represents a singular value, an original space-time matrix is reconstructed by utilizing a first mode, accuracy verification of a reconstruction result is carried out, and filling information of a missing point in the original matrix is updated by utilizing the reconstruction result;
d) repeating c) until convergence based on the updated matrix; and then, increasing the number of main modes, and repeating the steps until the overall situation converges, thereby realizing the missing information reconstruction in the current hourly scale AOD inversion product.
And step 3: the aerosol mode simulation results are corrected using a machine learning based data error correction technique, specifically:
a) extracting pairing data of the satellite remote sensing inversion AOD and the numerical mode simulation AOD at the corresponding moment;
b) establishing a random forest machine learning model by means of related auxiliary variables;
c) and the extrapolation model realizes the downscaling correction of the mode simulation AOD background field.
And 4, step 4: based on AERONET and SONET foundation actual measurement AOD data, national control air quality monitoring station actual measurement PM10，PM2.5，NO2And SO2The method comprises the following steps of (1) utilizing a random forest machine learning method to construct a day-hour resolution ratio AOD estimation model based on actually-measured pollutant concentration data and meteorological observation data MET at corresponding moment, and developing cross validation to evaluate the generalization capability of the model so as to estimate AOD data of a point position where an air quality monitoring station is located:
AOD～PM10+PM2.5+NO2+SO2+MET
and 5: the invention provides a method for performing satellite-foundation-mode multi-source heterogeneous data fusion analysis by using an optimal interpolation technology to produce high-precision full-coverage AOD product data, wherein a numerical simulation result is used as a background field of the fusion analysis, and satellite and foundation data are used as a fusion observation field.
For any pixel x to be analyzed, defining the background value of the pixel x to be xbThen it reanalyzes the value xaCan be expressed as:
xa＝xb+K(Yo-HXb)
wherein, YoAnd the observed value in the neighborhood of the pixel x to be analyzed. H is the observation operator, HXbRepresents the observed value YoThe corresponding background value. K is the kalman gain, which quantifies the effect of neighboring observations on the point to be analyzed and can be defined as:
K＝PbHT(HPbHT+R)-1
wherein HTA transposed matrix of H. The observation error covariance matrix R is a diagonal matrix, and diagonal elements are represented by corresponding observation error variances epsilonoComposition, background error covariance matrix PbIs a symmetric matrix, and can be expressed as:
wherein epsilonbTo approximate the background error variance, the error variance of the background field and the observation field can be estimated by the measured values, ρ (i, j) is used to represent the spatial correlation between the data points i, j, the present invention uses gaussian kernel weights to approximately fit this spatial correlation:
where dx and dy represent data points i, j, respectively, and the spatial distance dijThe orthogonal components in the warp and weft directions, lx and ly, represent the corresponding spatial window sizes.
In the specific fusion process, the space hierarchical clustering technology is adopted to resample the satellite inversion AOD data, and the problem of sample size imbalance between the satellite remote sensing inversion lattice point data and the ground station sparse observation data is solved. Specifically, a deviation field between a background field and an observation field is calculated, lattice observation values with an error not exceeding AOD +/-0.05 in a certain range (50km) are grouped into one type, and finally, a clustering result is input into a data fusion process.
By utilizing the fusion technology, the AOD satellite inversion and missing value reconstruction result, the state control point AOD estimation data and the AOD numerical prediction field are fused to generate a daytime-space seamless hourly resolution AOD data product
And 6: and (5) establishing a statistical relationship model based on data driving to estimate the surface area PM by combining the high-precision full-coverage AOD data generated in the step 52.5The density grid data.
a) Selecting other modeling variables except AOD to construct an auxiliary model, such as a meteorological variable MET mainly based on relative humidity and boundary layer height, a social and economic variable SE mainly based on population density and road network density, a related variable TIME reflecting TIME characteristics and the like;
b) according to the longitude and latitude positions of the sites, the variable in a) and PM are combined2.5Matching concentration observation data, extracting to obtain a two-dimensional training data set, and randomly reserving 10-20% of training data as a verification set for cross-checking model estimation accuracy for verifying the generalization capability of the model;
c) inputting the training data set into a random forest machine learning model for model parameter training:
PM2.5～AOD+MET+SE+TIME
d) developing modeling variable sensitivity analysis, gradually eliminating variables which do not significantly contribute to modeling precision by calculating the relative importance of input variables, and optimizing a model structure;
e) estimating the PM of the surface area by using the produced AOD fusion product based on an optimization model2.5Concentration grid point data.
And 7: fully excavating PM2.5Time autocorrelation characteristic of concentration sequence, and construction of PM at different time intervals2.5Concentration migration model to realize PM under the condition of non-satellite inversion AOD data at night2.5And (5) producing concentration seamless lattice point data.
a) According to the longitude and latitude information of the national control site, meteorological variables such as humidity, boundary layer height and the like and PM (particulate matter)2.5The concentration observation data are matched to obtain the PM of different sites2.5-MET observation time series set;
b) adopting a cross verification method, randomly selecting 80% of matched data samples as a training set, and 20% of matched data samples as a verification set;
c) with the current time t as a reference, forward-pushing m time observation time sequences to input an LSTM-FC layer to mine the time dependency:
d) performing model parameter training by adopting a gradient descent method, performing model precision evaluation by using a verification set, and performing model performance optimization by continuously adjusting super parameters such as learning rate;
e) the PM generated in the step 6 is treated2.5The concentration seamless lattice point data and the meteorological field data at the corresponding moment are input into a migration model established based on site observation together so as to realize PM under the condition of non-satellite inversion AOD data2.5And (5) producing concentration seamless lattice point data.
And 8: this step is intended to be directed to the PM produced in step 6 and step 72.5And performing optimization correction on the concentration seamless grid data.
a) Calculating the Euclidean distance d between each grid point and a national control station;
b) defining a search radius r, a relevant distance cl and the maximum station number n;
c) searching the nearest n national control sites within the radius r for any grid point;
d) defining the correction strength w of the state control station to the lattice values：
e) The correction value of the grid point
Can be expressed as:
wherein PMoriginalIs the original value of the grid point,
is the measured value of the domestic site in the neighborhood,
the original value of the grid point of the corresponding position is obtained;
f) circularly traversing each grid point to complete the whole PM2.5And optimally correcting the density image.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention are equivalent to or changed within the technical scope of the present invention.
Claims (9)
1. All-time PM2.5The near real-time production method of the concentration seamless lattice point data is characterized by comprising the following steps of:
step 1: rapidly inverting day hour-by-hour AOD data by depending on a remote sensing observation platform of a static satellite;
step 2: reconstructing missing information in the AOD lattice point data produced in the step 1;
and step 3: performing spatial downscaling and error correction on the AOD result of the numerical mode simulation based on satellite inversion AOD data by using a machine learning algorithm;
and 4, step 4: estimating the AOD data level of the corresponding point by utilizing ground actual measurement atmospheric particulate matter concentration data in combination with a random forest machine learning modeling method;
and 5: taking the reduced scale numerical simulation AOD result obtained in the step 3 as a background field, and fusing the multi-source AOD data obtained in the step 2 and the step 4 to obtain a daytime high-precision spatial seamless AOD product;
step 6: utilizing the seamless AOD product obtained in the step 5 and combining with the ground state control station to actually measure PM2.5Concentration, real-time meteorological observation data, social and economic data and time variable related to atmospheric pollution are modeled by adopting a random forest machine learning method to estimate surface area PM2.5Concentration distribution data;
and 7: aiming at the condition that no effective satellite remote sensing inverses AOD data at night and the like, PM between different moments is constructed2.5Concentration migration model for realizing PM of other time areas based on existing data of adjacent time2.5Estimating concentration data;
and 8: developing actual measurement PM of site at current moment2.5Concentration data and space-time seamless PM obtained in step 6 or 72.5Integration between lattice data products, high-precision full-coverage PM at current production time2.5The density grid data.
2. The all-time PM of claim 12.5The near real-time production method of the concentration seamless lattice point data is characterized in that the step 1 depends on the high time-frequency observation characteristic of a static satellite, and a minimum spectrum regression coefficient method is adopted to estimate the surface reflectivity so as to realize accurate ground gas decoupling.
3. The all-time PM of claim 12.5Number of concentration seamless latticeThe near real-time production method is characterized in that in the step 1, partial missing data reconstruction is realized by fusing multi-temporal incomplete AOD data by an empirical orthogonal function method according to the time-space autocorrelation characteristics of the AOD data, and the spatial coverage level of the satellite inversion AOD original data obtained in the step 1 is improved.
4. The all-time PM of claim 12.5The near real-time production method of the concentration seamless lattice point data is characterized in that the data reconstruction technology adopted in the step 2 is specifically described as introducing space-time neighborhood AOD data, constructing an AOD space-time correlation matrix at the current moment, performing iterative decomposition on the AOD space-time correlation matrix by using a Singular Value Decomposition (SVD) method, and iteratively reconstructing missing values in the space-time correlation matrix by using a main mode until convergence.
5. The all-time PM of claim 12.5The near real-time production method of the concentration seamless lattice point data is characterized in that the real-time air pollutant concentration provided by the national control air quality observation network is provided by ground actual measurement atmospheric particulate matter concentration data in the step 4, and the statistical modeling formula for estimating the AOD in the step 4 is as follows:
AOD～PM10+PM2.5+NO2+SO2+MET
wherein the AOD data is provided by an atmospheric aerosol characterization observation site, PM10、PM2.5、NO2And SO2The concentration data is provided by a national foundation air quality monitoring network, and MET represents meteorological observation data at a corresponding moment.
6. An all-time PM of claim 12.5The near-real-time production method of the concentration seamless lattice point data is characterized in that in the step 5, the optimal interpolation fusion technology is utilized to carry out fusion analysis on the multi-source heterogeneous AOD data:
xa＝xb+K(Yo-HXb)
wherein x isaAs a reanalyzed value, xbIs a background valueK is Kalman gain, YoIs the observed value in the neighborhood of the pixel x to be analyzed, H is the observation operator, HXbRepresents an observed value YoIn the invention, the reduced-scale mode simulation result is used as a fusion background field, and the satellite inversion result and the ground estimation data are used as a fusion observation field.
7. The full-time full-coverage PM of claim 12.5The near real-time production technical scheme of the concentration grid point data is characterized in that systematic error correction of AOD products with different data sources is represented by corresponding error variances of a background field and an observation field:
ε＝var(X-Y)
wherein, X is actually measured data, and Y is data to be calibrated.
8. An all-time PM of claim 12.5The near real-time production method of the concentration seamless lattice point data is characterized in that the modeling formula in the step 6 is as follows:
PM2.5～AOD+MET+SE+TIME
wherein PM2.5The data is provided by a national foundation air quality monitoring network, the AOD uses the space-TIME full coverage reanalysis data produced in the step 5, MET represents meteorological observation data at a corresponding moment, SE represents social and economic data related to atmospheric pollution, and TIME represents a corresponding TIME variable.
9. The all-time PM of claim 12.5The near real-time production method of the concentration seamless lattice point data is characterized in that PM between different moments in the step 72.5The concentration migration model is constructed by a deep neural network learning model, and the expression of the deep neural network learning model is as follows:
wherein t represents the current time, and t-m represents the base of t timeQuasi-forward-push m-time, LSTM is a long-short term memory layer, FC is a full-connection layer, a migration model is constructed based on site continuous actual measurement data, and model extrapolation is applied to realize surface domain size PM2.5And (4) estimating the concentration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111625657.3A CN114707396A (en) | 2021-12-28 | 2021-12-28 | All-time PM2.5Near real-time production method of concentration seamless lattice point data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111625657.3A CN114707396A (en) | 2021-12-28 | 2021-12-28 | All-time PM2.5Near real-time production method of concentration seamless lattice point data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114707396A true CN114707396A (en) | 2022-07-05 |
Family
ID=82166521
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111625657.3A Pending CN114707396A (en) | 2021-12-28 | 2021-12-28 | All-time PM2.5Near real-time production method of concentration seamless lattice point data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114707396A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115081557A (en) * | 2022-08-22 | 2022-09-20 | 北华航天工业学院 | Night aerosol optical thickness estimation method and system based on ground monitoring data |
| CN116486931A (en) * | 2023-06-21 | 2023-07-25 | 上海航天空间技术有限公司 | Full-coverage atmospheric methane concentration data production method and system coupled with physical mechanism |
-
2021
- 2021-12-28 CN CN202111625657.3A patent/CN114707396A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115081557A (en) * | 2022-08-22 | 2022-09-20 | 北华航天工业学院 | Night aerosol optical thickness estimation method and system based on ground monitoring data |
| CN116486931A (en) * | 2023-06-21 | 2023-07-25 | 上海航天空间技术有限公司 | Full-coverage atmospheric methane concentration data production method and system coupled with physical mechanism |
| CN116486931B (en) * | 2023-06-21 | 2023-08-29 | 上海航天空间技术有限公司 | Full-coverage atmospheric methane concentration data production method and system coupled with physical mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zhao et al. | A spatial downscaling approach for the SMAP passive surface soil moisture product using random forest regression | |
| Yang et al. | Filling the missing data gaps of daily MODIS AOD using spatiotemporal interpolation | |
| Boiyo et al. | Spatial variations and trends in AOD climatology over East Africa during 2002–2016: a comparative study using three satellite data sets | |
| Zeng et al. | Daily global solar radiation in China estimated from high‐density meteorological observations: a random forest model framework | |
| Tooke et al. | Integrated irradiance modelling in the urban environment based on remotely sensed data | |
| Song et al. | An improved surface soil moisture downscaling approach over cloudy areas based on geographically weighted regression | |
| CN114019579B (en) | High space-time resolution near-surface air temperature reconstruction method, system and equipment | |
| Chen et al. | Geometry and adjacency effects in urban land surface temperature retrieval from high-spatial-resolution thermal infrared images | |
| CN114707396A (en) | All-time PM2.5Near real-time production method of concentration seamless lattice point data | |
| Heo et al. | Digital elevation model-based convolutional neural network modeling for searching of high solar energy regions | |
| Larue et al. | Simulation and assimilation of passive microwave data using a snowpack model coupled to a calibrated radiative transfer model over northeastern Canada | |
| Yoo et al. | Spatial downscaling of MODIS land surface temperature: Recent research trends, challenges, and future directions | |
| Bennouna et al. | Atmospheric turbidity determined by the annual cycle of the aerosol optical depth over north-center Spain from ground (AERONET) and satellite (MODIS) | |
| Shwetha et al. | Prediction of land surface temperature under cloudy conditions using microwave remote sensing and ANN | |
| Fang et al. | A western United States snow reanalysis dataset over the Landsat era from water years 1985 to 2021 | |
| CN112016696A (en) | PM integrating satellite observation and ground observation1Concentration inversion method and system | |
| Xu et al. | A new algorithm for Himawari-8 aerosol optical depth retrieval by integrating regional PM₂. ₅ concentrations | |
| Lee et al. | Analytical approach to estimating aerosol extinction and visibility from satellite observations | |
| Xiong et al. | Estimating the PM2. 5 concentration over Anhui Province, China, using the Himawari-8 AOD and a GAM/BME model | |
| Tapiador et al. | Precipitation estimates for hydroelectricity | |
| Dehshiri et al. | Dust cycle, soiling effect and optimum cleaning schedule for PV modules in Iran: A long-term multi-criteria analysis | |
| Leng et al. | Determination of all-sky surface soil moisture at fine spatial resolution synergistically using optical/thermal infrared and microwave measurements | |
| Hong et al. | Assimilating Fengyun-4A observations to improve WRF-Chem PM2. 5 predictions in China | |
| Mushore et al. | Estimating urban LST using multiple remotely sensed spectral indices and elevation retrievals | |
| Han et al. | Estimation of high-resolution PM2. 5 concentrations based on gap-filling aerosol optical depth using gradient boosting model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |